Speech

00Overview

Turn audio and text into usable content straight from a flow - transcribe recorded audio files into text, synthesize spoken audio from written text, and train custom vocabularies that boost transcription accuracy for your own product names and jargon. Start transcription jobs against audio files in Object Storage, list the text-to-speech voices available for synthesis, and manage jobs and custom vocabularies end to end - fetch, list and update them, cancel or move a transcription job, and delete a customization. Everything runs against your own Oracle Cloud tenancy, in the region you choose.

Every field below is exactly what you see in the Flomation editor. Fields marked ● live picker let you choose from a list pulled live from your account — no IDs to look up.

01Connecting Speech

1. Choose the Authentication method. Connect Oracle Cloud uses a saved Oracle Cloud connection — pick it in the Oracle Cloud connection field and the signing-key details are supplied for you. API signing key (advanced) lets you paste the raw credential fields directly, described below.
2. For an API signing key, sign in to the OCI Console, open the Profile menu (top right) → My profile, then under Resources choose API keys → Add API key and either generate a new key pair or upload your own public key.
3. After the key is added, OCI shows a Configuration file preview — copy the Tenancy OCID, User OCID, Key Fingerprint and Region from it into the matching node fields (the region is a plain identifier such as uk-london-1).
4. Set Compartment OCID to the compartment that holds (or will hold) your Speech resources — find it under Identity → Compartments in the console. Leave Private Key Passphrase blank unless the key is encrypted.
5. Store the API signing private key (the full PEM, including the BEGIN/END lines) as a Flomation environment secret (e.g. speech_secret) and pick it in the node's Private Key (PEM) field rather than pasting it inline.

02Customization

OCI Speech: Create Customization

oracle/speech/customization_create · Action

Create a Speech customization — a custom vocabulary trained from files in an Object Storage bucket that boosts transcription accuracy for your own terms. Returns the customization in a CREATING state; poll Get Customization until ACTIVE.

Returns: tool_result, customization, id, lifecycle_state, success, error

OCI Speech: Delete Customization

oracle/speech/customization_delete · Action

Delete an OCI Speech customization by its OCID — it can no longer be used by transcription jobs.

Returns: tool_result, id, success, error

OCI Speech: Get Customization

oracle/speech/customization_get · Action

Fetch a single Speech customization by its OCID — its display name, alias, description and lifecycle state.

Returns: tool_result, customization, id, lifecycle_state, success, error

OCI Speech: List Customizations

oracle/speech/customization_list · Action

List the Speech customizations in a compartment. Optionally filter by exact display name or lifecycle state, and cap the page size. Walks pagination up to a safe cap.

Returns: tool_result, customizations, count, truncated, success, error

OCI Speech: Update Customization

oracle/speech/customization_update · Action

Partially update a Speech customization — change only the display name or description you supply; blank fields are left unchanged.

Returns: tool_result, customization, id, success, error

03List

OCI Speech: List Voices

oracle/speech/list_voices · Action

List the text-to-speech voices available for synthesis. Optionally filter by TTS model, language code, or display name.

Returns: tool_result, voices, count, success, error

04Synthesize

OCI Speech: Synthesize Speech

oracle/speech/synthesize_speech · Action

Convert text into spoken audio with the OCI Speech text-to-speech service and return the generated audio as base64.

Returns: tool_result, audio_base64, byte_count, success, error

05Transcription

OCI Speech: Cancel Transcription Job

oracle/speech/transcription_job_cancel · Action

Cancel an in-progress Speech transcription job by its OCID — any outstanding transcription tasks are stopped.

Returns: tool_result, id, success, error

OCI Speech: Change Transcription Job Compartment

oracle/speech/transcription_job_change_compartment · Action

Move a Speech transcription job into a different compartment — the job keeps its OCID, only its compartment placement changes.

Returns: tool_result, id, destination_compartment_id, success, error

OCI Speech: Create Transcription Job

oracle/speech/transcription_job_create · Action

Start a transcription job that converts an audio file in Object Storage into text using the chosen language model, writing the results to an output bucket. Returns the job in an ACCEPTED state — poll Get Transcription Job until SUCCEEDED.

Returns: tool_result, transcription_job, id, lifecycle_state, success, error

OCI Speech: Get Transcription Job

oracle/speech/transcription_job_get · Action

Fetch a single Speech transcription job by its OCID — its lifecycle state and percent complete.

Returns: tool_result, transcription_job, id, lifecycle_state, percent_complete, success, error

OCI Speech: List Transcription Jobs

oracle/speech/transcription_job_list · Action

List the audio transcription jobs in a compartment. Optionally filter by exact display name or lifecycle state, and cap the number returned. Walks pagination up to a safe cap.

Returns: tool_result, transcription_jobs, count, truncated, success, error

OCI Speech: Update Transcription Job

oracle/speech/transcription_job_update · Action

Partially update a Speech transcription job — change only the display name or description you supply; blank fields are left unchanged.

Returns: tool_result, transcription_job, id, success, error

06Notes & Limitations

Behaviours and constraints worth knowing before you build with these nodes.

• Transcription jobs and customizations both run asynchronously: Create Transcription Job returns a job in the ACCEPTED state and Create Customization returns a customization in the CREATING state, so poll the matching Get action until the job reaches SUCCEEDED or the customization reaches ACTIVE before using the results.
• The language code on a transcription job must match the model type you choose: the Oracle model expects locale codes such as en-US or fr-FR, while the Whisper models expect short codes such as en or es, and a mismatched pairing is rejected.
• A transcription job reads its audio from an Object Storage bucket and writes the results into an output bucket, and a customization trains from an object in a bucket, so those buckets and objects must already exist and the Speech service must be granted access to them before the job runs.
• Synthesize Speech has no voice selector and always uses the service's default text-to-speech voice and audio format, so List Voices is informational only and does not change what Synthesize Speech produces.
• The list actions walk pagination only up to a fixed page limit, so in a large compartment apply the display-name or lifecycle-state filter to be sure the resource you need is returned.